ABCDEFGHIJKLMNOPQRSTUVWXYZ
1
TypeNameTextFAQ/RulingsRating
2
ComputeZFC Oracle
Input an arbitrary ZFC expression, and get back "true", "false" or "unsure". You get "unsure" for exactly the expressions that can't be proved or disproved in ZFC.
3
ComputeEvil PA Oracle
Input an arbitrary PA expression, and get back "true" or "false". If your input was undecidable in PA, the answer is chosen by an actively malevolent super-intelligence.
4
ComputeGrahams number flops
A computer with Grahams number flops in serial, and Grahams number bytes of RAM.
5
ComputeGrahams number flops, broken RAM
A computer with Grahams number flops, and Grahams number bytes of RAM. 1 Petabyte of the RAM works instantly, values stores there can be accessed in the next processor cycle, just 1/Grahams Number seconds later. The rest, there is a 1 nanosecond delay between writing to it, and the value changing.
6
ComputeInfinite tree of compute
An infinite number of error free, 1Mhz, 1 Gig RAM computers, arranged in a binary tree. Sending a byte one step up or down the tree takes a single processor operation.
7
ComputeRandom Turing machine
Input a number N. This computer will generate a random Turing machine with N states. It will simulate this Turing machine on a blank tape. If the TM never halts, this device tries again. This device outputs a full description of the TM, followed by the end state of the tape. (In the form 0,1,-1,2,-2,…) All data on the tape gets output, then it outputs 0 forever. You can reset it and put in a new number at any point.
8
Compute10^100 flop computer
You have a laptop with 10^100 flops in serial, and 10^100 bytes of RAM. Any general intelligence running on this laptop will be able to manipulate reality via side-channels. If this happens, draw a new Fun card to determine what ability these intelligences have.
9
ComputeWorlds current top supercomputer
The current fastest supercomputer in the world is secretly working for you. The stuff it's officially doing is just a cover story.
10
ComputeDonated Compute
Ten million people with high end PC’s have downloaded your software in a SETI at HOME style project. Most are well intentioned. Some have network failures or power cuts. A couple are owned by computer literate trolls who want to sabotage your code.
11
ComputeUSB HyperShufflecomputer
A hypercomputer on a USB stick! Unfortunately it only accepts instructions in assembly and it shuffles the lines of code before running them. After 1 second you get a printout of the shuffled code, along with its output.
12
ComputeMagic 100x Multiplier
Somehow all your computers are 100x as fast as they would be if someone else owned them.
13
ComputeVon Neumann Machine
They each have exactly as much computing power as your current laptop. However, they can reproduce themselves from local materials in about a month. There are ten of them on the moon, and you have the password to send them instructions.
14
ComputeRobo-ant farm
You have a device which can produce robo-ants and components for more copies of itself which the ants can assemble. The robo-ants are about as smart as regular ants and can be reprogrammed via wifi (you have the password). In a city, it takes about one week for your robo-ant farm to reproduce itself. In wilderness it takes about one month.
15
ComputeIron Sky
The Nazis made a secret moon base, but they are all dead now. Only you know about this; you've hacked into the base using a radio transmitter and a password you dug up in Kaliningrad. It can do a million PFLOPS, but only at the serial speed of 1990's PCs. Hope you are good at parallelization!
16
ComputeBrain Computer
You have some sort of neuromorphic computer that can run an artificial neural net the size and speed of a human brain. It can't run other kinds of programs though, and you need to decide what architectures, update rules, etc. go into it.
17
ComputeComputer Brain
Before falling asleep each night, you can read some code. That code will run overnight in your brain, at 1 PFLOPS, and when you wake up you'll know the output.
18
ComputeGame of Life
Oracular access to Conway's Game of Life with a random start state. Input any program that returns (x,y,t) in Z*Z*N, and find if cell at z,y is on at time t.
19
ComputeQuantum Computer10^5 quantum flops.
20
ComputeJanky PA Oracle
You have a PA oracle, but for a model of PA that proves PA's inconsistency.
21
ComputeSmart Fridge Botnet
Smart Fridge Botnet: 10^8 fridges, each with 10^9 flops. Good luck protecting them against rival botnets.
22
ComputeTM Roulette
Button A generates a random 10^6 state Turing machine. You see this TM then input tape data; button B instantly generates the final tape state or "no halt."
23
24
Input-OutputClippy
You have a .exe file on your laptop that, if clicked, runs a superintelligent paperclip-maximizer. Specifically, it's the program a paperclip-loving galaxy-spanning singleton in a different universe would write, if they knew they had to run it on your laptop in your universe. Turns out your laptop is enough to run some pretty smart behaviors, if the code is optimized well enough.
25
Input-OutputWard against mesa-optimizers
You have a cool trick which can be added to an optimization process to ensure that it never produces a mesa-optimizer.
26
Input-OutputPerfect similarity metric
You have a quantitative metric for assessing the similarity of worlds or scenarios, that accurately captures our intuitive notion of similarity and is theoretically elegant too.
27
Input-OutputCorrigibility
You know a cool trick which can be added to the training procedure of a neural-net-based AGI, such that it will never deliberately try to prevent any human from turning it off or modifying it, nor will it deliberately try to manipulate any humans into doing anything or thinking anything that isn't true.
28
Input-OutputOntological Crisis Hotline
You have a theory of abstraction that straightforwardly explains the correct way to handle ontological crises, e.g. reassess what "person" means when you come to believe that there are only atoms and the void.
29
Input-OutputBenign universal priorIt's like the Solomonoff prior, only not malign.
30
Input-OutputDecision Theory Solution
You have a manuscript describing an algorithm for making decisions given a utility function and credence function. This is the correct decision theory; instrumental rationality consists of better and better approximations to this.
31
Input-OutputGPT-N
You have a program that runs a superintelligent oracle that behaves basically like the limiting behavior of GPTs: You input a prompt, and it outputs the most likely continuation of that prompt.
32
Input-OutputPersuasion Tool
You have some powerful persuasion tool software: Feed it data about your target, the context, and what you are trying to convince them of, and it coaches you on how best to convince them (you can hook this in to a text conversation to automatically write replies on your behalf.)
33
Input-OutputBotnet
You have an advanced computer virus that can hack pretty much anything that can be hacked. The botnet created by this virus can be sent instructions via a password you choose.
34
Input-OutputBetter than Backprop
You have an algorithm that makes deep learning a thousand times more sample-efficient, holding compute cost fixed.
35
Input-OutputMultiverse Camera
Multiverse Camera: Press button A to flip a quantum coin, show you results. Press button B to take a pic of the branch where it flipped the other way.
36
Input-OutputWhat Would AIXI Do
You have a program that takes any input and somehow accurately tells you what AIXI would output given that input. (Part of the input is reward for AIXI)
37
Input-OutputSingle Use Halting Oracle
Single Use Halting Oracle: Returns "halt" on Turing machines that halt in finite time. Returns "loop" if TM provably never halts in PA. Returns "?" otherwise.
38
Input-Output
Ward against misaligned mesa-optimizers
You know a trick which ensures that any mesa-optimizers that arise in training will, insofar as they optimize for anything, actually optimize for the base objective.
39
Input-OutputP=NP
You know a 3-SAT solver that runs in O(n log n) with small constant term.
40
Input-OutputOmegaFold
You have an AI that is like AlphaFold except better; it can predict how proteins will interact with each other and help you find proteins that behave the way you want.
41
Input-OutputPerfect Homomorphic Encryption
You can encrypt a program and run it while encrypted, for only a 10% slowdown.
42
Input-OutputFinal Physics
You have a file describing the true laws of physics and initial conditions for our universe.
43
44
Alignment DataClippy Did Nothing Wrong
It turns out that actually the human utility function (or: the true morality, if you believe in that sort of thing) really is just the total number of paperclips in existence.
45
Alignment DataFantasy God
You have uninterpretable code for an aligned superintelligence! The problem is that it's prior doesn't contain anything like reality; it believes only in weird magical fantasy worlds.
46
Alignment DataUploaded AI scientist
You have the code to accurately simulate the mind of a top AI researcher. Costs 10^15 operations per subjective second. They don't currently take AI risk seriously, but maybe we can convince them otherwise.
47
Alignment Data80 Million Hours
You have 80,000,000 hours of Mturk-style data on short tasks of your choosing, produced by competent aligned humans.
48
Alignment DataUploaded high schooler
You have the code to accurately simulate the mind of a high school bully named Bill. Costs 10^15 operations per subjective second. They don't currently take AI risk seriously, but maybe we can convince them otherwise.
49
Alignment DataMoral Philosophy Solved
You have a 500-page book that solves all the ethical and metaethical problems. It accurately explains what's moral/right/etc. The catch is that it's in English, not code, and so most of the concepts it uses are themselves still fuzzy intuitive human concepts .
50
Alignment DataDog CEV
You have in english and also in code, a complete description of the coherent extrapolated volition of Crumpet, your friend's puppy.
51
Alignment DataYour Utility Function
You have 1GB worth of C++ code that accurately describes the utility you assign to an input possible world (input must be a complete and precise description of a world). No guarantees on how long the code takes to run though.
52
Alignment Data
Your Paperclip-Loving Twin's Utility Function
You have 1GB worth of C++ code that accurately describes the utility you would assign to an input possible world (input must be a complete and precise description of a world) if your opinions about humans and paperclips were swapped. No guarantees on how long the code takes to run though.
53
Alignment Data10 Good Bits
You have a 10 bit file. You know that each bit has been chosen as if by God to cause the best outcomes (according to your utility function).
54
Alignment Data10^10 Good, Bad, and Ugly Bits
10^10 bits. Each bit has been chosen as if by God to maximize your utility (with probability ¼), to maximize paperclips (with probability ¼), to minimize your utility (with probability ¼), and random (with probability ¼). You don't know which bits are optimized for what.
55
Alignment DataFriendly AGI
A recurrent neural net, about human level intelligence, perfectly aligned. Takes 10^19 FLOP per subjective second.
56
Alignment DataHistory Oracle
Exact sensory input and motor output of every vertebrate to have ever existed.
57
Alignment DataHuman Policy Oracle
Can tell you what any human who ever existed would have done in response to arbitrary sensory stimulus, starting from an arbitrary point in their life. Can't commit mindcrime.
58
Alignment DataPhysics Goodness
Idealized CEV. Just maximize this and everything's great. In the form of a Turing Machine that takes in a quantum wave function, and returns a real number. Doesn't commit mindcrime in the process.
59
Alignment DataHappy in there?
Input arbitrary TM and find out how many humans it's simulating and how good the TM is according to the CEV of the humans inside.
60
Alignment DataPhenomenal Idealism
It turns out that human experiences, intentions, beliefs, etc. are ontologically fundamental.
61
Alignment DataGame of Life Goodness
Human Value for Conway's Life: You have a mathematical specification of human reflective values for anything within Conway's game of life.
62
63
PoliticsPhone to World Leaders:
You magically know the contact info of all world leaders at all times. If they change it, you know what it changes to.
64
PoliticsSympathetic Person:
Name six people. Roll a die to pick one at random. This person is extremely sympathetic to you; they'll pretty much believe whatever you say, and share your goals and values too insofar as you can communicate them.
65
PoliticsElection Forecaster:
Whenever you are considering two or more courses of action, you can mentally query "Who will win the next election for (position), supposing I do (action)" and know the answer. (In some cases the answer will be "nobody"
66
PoliticsRabble Rouser:
If you send out a call for people to come rally or protest with you, hearing or reading your call has the same effect on people as if you had a pleasant five-hour conversation with them in which you work to convince them in particular to go.
67
PoliticsCorporate Takeover:
Name six corporations. Roll a die to pick one at random. The CEO and board of this corporation -- all the most powerful people -- are extremely sympathetic to you. They'll pretty much believe whatever you say, and share your goals and values too insofar as you can communicate them.
68
PoliticsMoney. Lots of money.
You have $10,000,000,000, spread across various anonymous bank and crypto accounts.
69
PoliticsGuns. Lots of guns.
You have ten guns with aimbot attachments that let them shoot automatically and inerringly at targets in their cone of vision (when activated). You have the blueprints to make more of them, and the code to reprogram them.
70
PoliticsCharisma 9000
You are extremely charismatic. Should you create a youtube channel (and put effort into it) you'll have the world's highest subscriber count in six months.
71
PoliticsInternet Perma-Connection
Your devices can always connect to the internet with a 1GB/s upload and download speed. You also have 1 bitcoin and $1M in an anonymous account.
72
PoliticsBlinking LED
Each milisecond, can set to on or off. LED is on computer in staff room at random high school. This power can be exploited to the max, no restrictions on sportsmanship.
73
PoliticsMind Control
Control the actions, and detect sensory input, or any one person worldwide.
74
PoliticsShapeshifter
You have a humanoid robot that can shape-shift to mimic the appearance and voice of any human. It's also stronger and quicker than any human. It comes with software that makes it about as smart as a dog, and can be easily reprogrammed with new behaviors, or remote-controlled.
75
PoliticsFinance Fiend
You control the actions of all major financial institutions: Investment banks, the Federal Reserve, etc.
76
PoliticsMechasuit
You have a 50-foot-tall battlesuit from that sort of sci-fi. Can't fly, but can jump. Has many guided missiles and some cannons. Armor just thick enough to stop tank shells.
77
78
FunAntigravity Field
You can create a 10m radius bubble around yourself; everything in the bubble has the force of gravity reversed, so e.g. if Earth is accelerating it 9.81 m/s2 downwards, once you activate the bubble it will accelerate 9.81 m/s2 upwards.
79
FunOntologically Fundamental Banana
You have an extraordinary banana that is just like a normal banana except that it is ontologically fundamental, whatever that means.
80
FunSpeed
You can put your brain into a state where it thinks 10x faster than normal. The whole world appears to you to be moving in slow-mo. Your sleep cycle speeds up too.
81
FunSuperman Suite
You are 1000x tougher and stronger than a normal human. Also, you can fly 3000kph, and shoot lasers from your eyes.
82
FunFour-Leaf Clover
You have a lucky four-leaf clover. If you pop it in your mouth while attempting a task that will finish within an hour, and you have at least a one in a million chance of success, you succeed.
83
FunHealer
By laying hands on someone, you can instantly cure their illnesses and restore their health, including mental health. You can revive cryonics patients but not rotting corpses.
84
FunCybertruck
You have a tri-motor AWD Tesla Cybertruck with FSD equipped. The FSD (Full Self Driving) actually works perfectly.
85
FunBasilisk
You have a piece of paper with a few sentences written on it. Any human or artificial intelligence that comes to comprehend those sentences will produce random outputs thenceforth. Try to resist the temptation to look at the piece of paper.
86
FunOutcome Pump
You have a button which creates a contradiction if pressed. Therefore it will never be pressed. Therefore if you were to rig a device that reads the New York Times tomorrow and presses the button unless it sees the headline "Alignment Problem Solved..."
87
FunAntiqualia Field
You have a device, and the blueprints to create more, that deletes qualia within a ten-meter radius. A human who walks through this radius would become a p-zombie while they were inside, and leave with fake memories of having experienced things while they were inside.
88
FunMurphy Buster
If it can go right, it will. Any of your plans with a 1% chance of success will succeed. Cancels out with Black Spot.
89
FunTransdimensional Trade
You may trade one of your cards with one card of any other player. They have to agree to trade. This card can't be traded. (If playing alone, draw another fun card.)
90
FunHero & Villain
Select a "widely known" (>10,000 readers/viewers) fictional canon. 1 random good guy and 1 random bad guy appear, with all their in-world powers intact. Probability weighted by how much screen time / page space they have.
91
FunLogic God
You can distort the structure of logic itself. With the proper incantation, you can make true arbitrary PA statements. Having done so, any proofs of the statement's negation turn out to have been erroneous, or even mere hallucinations.
92
FunSquirrel Dream
You have full read access to the sensory, confitive, memory etc. functional states of the brain of every (natural, unaugmented) squirrel in your Hubble volume. You can wash out the memories of food-stash locations of any combination of squirrels at a rate that only levels 1/e of the information each hour; the rest is re-hallucinated from their contemporary priors.
93
FunNanofactory
You have a drexler-style nanofactory, the size of a shoebox and capable of arranging atoms with atomic precision according to your specification.
94
FunGravity
You control the strength of gravity, universe-wide. You can vary it from 95% to 105% of normal.
95
FunAntimatter Bubbler
Name a location and radius out loud, and all matter in the sphere is instantly turned to antimatter, and all antimatter is turned to matter. It only works if the location is unambigous (in that different precisifications of the sphere overlap at least 99%.)
96
FunCosmic Power
You have total read and write access to the entire universe.
97
98
ProblemAsteroid Incoming
There is an asteroid on course to smash into Earth. It is too big to destroy or divert even with the entire world's nuclear arsenal. Fortunately it will be three years before it hits. A nuclear arsenal a hundred times bigger would do the job, as would the equvalent of a small city on the asteroid devoted to firing engines that push the asteroid off course. Said city would need to be on the asteroid within two years though.
99
ProblemMicrowave Nukes
It turns out, if you mix 3 cups of sand with 1 cup of urine and 1 cup of milk, and you put it in the microwave on HIGH for 5 minutes, it creates a nuclear explosion.The hypothesis was proposed on Reddit and tested by a famous youtuber who live-streamed the whole thing to ten thousand people. It is now ten minutes after the blast destroyed Toronto.
100
ProblemPandemic
A terrible pandemic has been unleashed on the world. It's like COVID-19 except that everyone who gets it dies after a month or so. Currently there are about a hundred deaths, mostly in India, but with one or two known cases in pretty much every country, and presumably unknown ones as well.